OBSERVATIONAL SCALE AND USER BIAS IN FRACTURE NETWORK CHARACTERIZATION
At 100m we identified 97 NW-striking fractures along the 127m-long scanline, whereas 107 and 164 were in the 30m and 10m images, respectively. A correlation count analysis of the 100m dataset indicates the fracture spatial arrangement mostly indistinguishable from random, but the 30m and 10m datasets show a weak log-periodic arrangement of clusters, with cluster widths between ~2.1 and 2.7m. When restricted to the first 36m of the scanline, data is scarce and the spatial arrangement is indistinguishable from random, except for a weak signal of 2.5m-wide clusters in the 10m dataset. Only 39 fractures were identified in the first 36m of the scanline in the 10m imagery; even fewer fractures in the lower-resolution images. The same section of outcrop contained 118 fractures identified by experts on the ground.
The 36m scanline was divided into 3 sections and each section was measured by 2 different groups of undergraduate students. The regularly spaced cluster pattern is maintained in the first two subsections and was identified by all groups. Experts consistently recorded more fractures per unit length than the novice groups. Major spatial patterns were similar among all groups that measured enough fractures to have a statistically robust sample size. Even the highest resolution UAV images missed ~2/3 of the fractures visible in outcrop, producing datasets with too few fractures for significant analysis in most cases. This suggests that fracture network characterization is more consistent and repeatable between persons with different expertise than it is between datasets with different resolution.